Adaptive task scheduling method based on multi-core dsp

Abstract

The invention relates to a self-adaptation task scheduling method based on a multi-core DSP. The self-adaptation task scheduling method includes the steps that common task pools need to be established in a multi-core DSP system firstly, all cores participating in the operation are scheduled through an algorithm, and corresponding tasks are taken away from the common task pools and are calculated. The tasks are an equalizing task, an increasing task, a decreasing task and a random task respectively. A self-adaptation scheduling process is achieved through three steps that firstly, the front parts of the whole tasks are evenly distributed to multiple processor cores to be processed, the average processing time of the task which each core is responsible for is calculated, and task types are speculated; secondly, corresponding scheduling strategies are selected according to the speculated task types until all the tasks in the task pools are taken away; thirdly, the cores which finish the tasks first find the task pools where the cores have the most surplus tasks so as to share workloads. According to the self-adaptation task scheduling method, self-adaptation scheduling can be performed according to features of different tasks, the tasks are evenly distributed to the processor cores as much as possible, and the overall computing performance is improved.

Description

technical field [0001] The invention relates to a task scheduling method in a multi-core DSP, belonging to the field of information technology. Background technique [0002] The research on automatic parallelization began in the 1970s, and has gone through several stages such as automatic vectorization, loop-level automatic parallelization, and process-level automatic parallelization. The development of each stage reflects people's continuous deepening of understanding of the parallel characteristics of programs. Through the discovery of the law of vectorized programming, people put forward the theory of dependency analysis, thus completing the automatic vectorization of serial programs. Following the successful application of autovectorization techniques, researchers turned their attention to automatic parallelization of shared memory architectures, with some success. However, how to distribute data to multiple processors to fully explore the parallelism of the program and...

AI Technical Summary

Problems solved by technology

Dynamic scheduling is to dynamically assign tasks to idle threads at runtime, but this will cost a certain amount of scheduling overhead

[0007] Most of the existing task scheduling algorithms need to confirm the task scheduling strategy during the coding process. Once it is determined that it cannot be modified, this will directly lead to a rigid and inflexible task scheduling strategy.

Images